| |
BGonline.org Forums
0-ply GNU Rollouts
Posted By: Chuck Bower In Response To: 0-ply GNU Rollouts (Camfella)
Date: Saturday, 23 August 2008, at 7:48 p.m.
My personal feelings (but shared by at least some others):
NEVER use truncated rollouts for cube decisions.
NEVER use 0-ply cube. (Use 2-ply.)
(Personally, I never use truncated anything, although I'm less adamant about truncated checkerplay.)
Prefer 2-ply checkerplay as well, but sometimes 0-ply is OK. Ironically 0-ply isn't that much faster than 2-ply when you consider joint-standard-deviation (which you should). The reason is that fewer 2-ply trials are required to attain the same JSD significance compared to 0-ply. Yes, 0-ply usually still gets to a particular JSD more quickly, but not in the proportion of square root of ratio of number of trials as one might expect.
Many (including me, occasionally) compromise with stepped rollouts (an even number of plies at 2-ply play followed by 0-ply play the remainder of the game). For most of the comparisons I've seen, these do as well as full 2-ply play.
Complicated (multiple checkers back) positions are traditionally where people go with full 2-ply, 2-ply exclusively (in my observation). And unfortunately these positions take the longest because of the many plays required to reach completion.
I still run muscular settings on a 1.6 GHz Core Solo. Faster computers are obviously better and I have some of those, but they're not always at my fingertips.
Patience is a virtue.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.